Recently I ve searching about Reinforcement deeplearning ,and I discoverd microsoft malmo project
malmo project provides some interface to control the actor in minecraft and get some feedback
tensorforce project provides Reinforcement learning api based on tensorflow
so my goal is to use tensorforce to play minecraft via malmo
In my opinion:
in tensorforce,I need to pass state to the agent ,and get the action from agent
action = agent.act(state)
state, terminal, reward = environment.execute(action)
then pass action to malmo (parse to command
) and get the state (maybe world_state.observations
)
agent_host.sendCommand(command)
world_state = agent_host.getWorldState()
obvsText = world_state.observations[-1].text
but how ? I dont know .and doing research now.
I really need some advice